Deep Learning for Mobile Multimedia

Ota, Kaoru; Dao, Minh-Son; Mezaris, Vasileios; Natale, Francesco G. B. De

doi:10.1145/3092831

Cited by 127 publications

(73 citation statements)

References 65 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Graphical Processing Unit (GPU) could play an important role in DL because of their parallel processing structure that speeds up both learning and inference (Ota et al, 2017). This application requires a host computer with an NVIDIA GPU card, where the proposed algorithms of the DNN are parallelized.…”

Section: Parallel Dnn Training-based Translatormentioning

confidence: 99%

An Enhanced Training- Based Arabic Sign Language Virtual Interpreter Using Parallel Recurrent Neural Networks

Abdou¹

2018

Journal of Computer Science

View full text Add to dashboard Cite

Intelligent machine translation systems have a remarkable importance in integrating people with disabilities in community. Arabic to Arabic sign language systems are limited. Deep Learning (DL) was successfully applied to problems related to music information retrieval, image recognition and text recognition, but its use in sign language recognition is rare. This paper introduces an automatic virtual translation system from Arabic language into Arabic Sign Language (ASL) via a popular DL architecture: The Recurrent Neural Network (RNN). The proposed system uses a deep neural network training-based system for ASL that convolves RNN and Graphical Processing Unit (GPU) parallel processors. The system is evaluated using both objective and subjective measures. Obtained results are towards reducing errors, speeding up avatar and expressing signs and facial expressions in a well-received manner by Deaf. The signing avatar is highly encouraged as a simulator for natural human signs.

show abstract

Section: Parallel Dnn Training-based Translatormentioning

confidence: 99%

An Enhanced Training- Based Arabic Sign Language Virtual Interpreter Using Parallel Recurrent Neural Networks

Abdou¹

2018

Journal of Computer Science

View full text Add to dashboard Cite

show abstract

“…After obtaining DNN models trained with a huge amount of data, they can be deployed for inference, perception and control tasks in various autonomous systems and internet-of-things (IoT) applications. Recently, along arXiv:2001.00138v4 [cs.LG] 22 Jan 2020 with the rapid emergence of high-end mobile devices 1 , executing DNNs on mobile platforms gains popularity and is quickly becoming the mainstream [9,28,30,43,63] for broad applications such as sensor nodes, wireless access points, smartphones, wearable devices, video streaming, augmented reality, robotics, unmanned vehicles, smart health devices, etc. [2,3,29,46,50].…”

Section: Introductionmentioning

confidence: 99%

PatDNN: Achieving Real-Time DNN Execution on Mobile Devices with Pattern-based Weight Pruning

Niu

Lin

et al. 2020

Proceedings of the Twenty-Fifth International Conference on Architectural Support for Programming Languages and Operating Syste

189

113

View full text Add to dashboard Cite

With the emergence of a spectrum of high-end mobile devices, many applications that formerly required desktop-level computation capability are being transferred to these devices. However, executing Deep Neural Networks (DNNs) inference is still challenging considering the high computation and storage demands, specifically, if real-time performance with high accuracy is needed. Weight pruning of DNNs is proposed, but existing schemes represent two extremes in the design space: non-structured pruning is fine-grained, accurate, but not hardware friendly; structured pruning is coarse-grained, hardware-efficient, but with higher accuracy loss.In this paper, we advance the state-of-the-art by introducing a new dimension, fine-grained pruning patterns inside the coarse-grained structures, revealing a previously unknown point in the design space. With the higher accuracy enabled by fine-grained pruning patterns, the unique insight is to use the compiler to re-gain and guarantee high hardware efficiency. In other words, our method achieves the best of both worlds, and is desirable across theory/algorithm, compiler, and hardware levels. The proposed PatDNN is an endto-end framework to efficiently execute DNN on mobile devices with the help of a novel model compression techniquepattern-based pruning based on an extended ADMM solution framework-and a set of thorough architecture-aware compiler/code generation-based optimizations, i.e., filter kernel reordering, compressed weight storage, register load redundancy elimination, and parameter auto-tuning. Evaluation results demonstrate that PatDNN outperforms three state-ofthe-art end-to-end DNN frameworks, TensorFlow Lite, TVM, and Alibaba Mobile Neural Network with speedup up to 44.5×, 11.4×, and 7.1×, respectively, with no accuracy compromise. Real-time inference of representative large-scale DNNs (e.g., VGG-16, ResNet-50) can be achieved using mobile devices.

show abstract

“…Addressing this ever-increasing demand of data hungry devices in an efficient and effective manner has driven the wireless industry to look into new paradigms. Device-to-device (D2D) and machine-to-machine (M2M) communications are viewed as promising solutions to this complex problem and hence, a key enabling technology for 5G IoT [16][17][18][19][20].…”

Section: Introductionmentioning

confidence: 99%

Construction Low Complexity and Low Delay CDS for Big Data Code Dissemination

et al. 2018

View full text Add to dashboard Cite

The diffusion of codes is an important processing technology for big data networks. In previous scheme, data analysis was conducted for small samples of big data and complex problems that cannot be processed by big data technology. Due to the limited capacity of intelligence device, a better method is to select a set of nodes (intelligence device) to form a connected dominating set (CDS) to save energy, and constructing CDS is proved to be a complete NP problem. However, it is a challenge to reduce the communication delay and complexity for urgent data transmission in big data. In this paper, an appropriate duty cycle control (ADCC) scheme is proposed to reduce communication delay and complexity while improving energy efficient in CDS-based WSNs. In ADCC scheme, the method for constructing CDS is proposed at lower complexity. Nodes in CDS are selected according to the degree of nodes. Then, duty cycle of dominator nodes in CDS is higher than that of dominated nodes, so the communication delay in the proposed scheme is far less than that of previous scheme. The duty cycle of dominated nodes is small to save energy. This is because the number of dominator nodes in CDS is far less than the number of dominated nodes whose duty cycle is small; thus, the total energy consumption of the network is less than that of the previous scheme. As a result, the performance of energy consumption and communication delay and complex have been improved. Its complexity O ∑ v i=0 m − i + 2m − 2v − s is reduced a lot for big data. The theoretical analysis shows that compared to the previous scheme, the transmission delay can be reduced 25-92% and the energy efficiency is improved by about 80% while retaining network lifetime.

show abstract

Deep Learning for Mobile Multimedia

Cited by 127 publications

References 65 publications

An Enhanced Training- Based Arabic Sign Language Virtual Interpreter Using Parallel Recurrent Neural Networks

An Enhanced Training- Based Arabic Sign Language Virtual Interpreter Using Parallel Recurrent Neural Networks

PatDNN: Achieving Real-Time DNN Execution on Mobile Devices with Pattern-based Weight Pruning

Construction Low Complexity and Low Delay CDS for Big Data Code Dissemination

Contact Info

Product

Resources

About